Evaluation of EEG Foundation Models for Event-Based Burst-Suppression Detection in ICU

2026-06-18

AI Summary

This study addresses the challenge of automatically detecting burst suppression (BS) patterns in intensive care unit (ICU) electroencephalography (EEG), which is hindered by high inter-patient heterogeneity and scarce annotations. For the first time, EEG foundation models—including REVE-base, LUNA-large, and LuMamba-Tiny—are applied to event-level burst detection in reduced-lead ICU EEG without requiring patient-specific calibration. To mitigate annotation variability, the authors introduce clinically motivated event-level evaluation metrics and systematically compare transfer learning strategies, including full fine-tuning, frozen backbone, two-stage fine-tuning, and LoRA. Experimental results show that fully fine-tuned REVE-base achieves an event-level F1 score of 0.868, reducing per-minute burst errors by 52.1% and 36.2% compared to EEGNet and adaptive thresholding, respectively. Notably, even with only 25% of the labeled data, it outperforms random initialization by 0.723 F1 points.

Abstract

Burst suppression (BS) is a clinically relevant electroencephalographic (EEG) pattern used to monitor sedation depth and brain activity in critically ill patients, particularly during induced coma in Intensive Care Units (ICUs). Automatic burst detection remains challenging because BS patterns vary substantially between patients and annotated datasets are scarce. Recently, EEG Foundation Models (FMs) have shown promise across several downstream EEG applications, but their usefulness for BS detection remains unexplored. We present the first study to evaluate EEG FMs for burst detection in reduced-montage ICU EEG without patient-specific calibration. We compare REVE-base, LUNA-large and LuMamba-Tiny with an adaptive thresholding baseline and a task-specific EEGNet baseline. Additionally, we complement conventional EEG window-based classification with event-based burst detection evaluation. This helps assessing clinically whether burst episodes are correctly detected, reducing the impact of expected annotation variability. The best model, REVE-base, achieved the highest event-based F1-score ($0.868 \pm 0.167$) and reduced burst-per-minute error by 52.1% and 36.2% compared to EEGNet and adaptive thresholding respectively, supporting FMs for scalable EEG monitoring in ICU. Ablation experiments showed that full fine-tuning was the most effective adaptation strategy with respect to frozen-backbone training, two-step fine-tuning, and LoRA-based adaptation, improving event-based F1-score over frozen-backbone training by up to $+0.102$ for LUNA-large. With reduced labeled datasets, pretrained REVE-base outperformed random initialization by $+0.723$ event-based F1 points at 25% of the cohort, demonstrating the benefit of pretraining FM representations when adapted to burst detection with limited labeled data.

Problem

• burst suppression
• EEG
• ICU
• automatic detection
• data scarcity

Innovation

• EEG Foundation Models
• burst-suppression detection
• event-based evaluation
• few-shot adaptation
• ICU EEG monitoring